EXCLUSIVE - PUE is a great starting point, but is it enough? – A discussion on government data centre efficiency
OpenGov held its third Breakfast Insight session on data centre efficiency on July 19 (reports from the 2015 and 2016 events), in partnership with Schneider Electric. Delegates from 14 government agencies and departments in Singapore attended, sharing their knowledge and asking difficult questions over the course of two hours of vibrant discussion.
Following opening remarks by Mr. Mohit Sagar, Managing Director and Editor In Chief of OpenGov Asia, Michael Kurniawan (below centre), Vice President, IT Division – Singapore, Schneider Electric kickstarted the discussion talking about the key role of ever increasing volumes of data and data centres in Singapore’s Smart Nation plans and the consequent need for efficient management of data centres.
Global trends in data centres
Guest speaker, Greg Boorer (below left), CEO, Canberra Data Centres (CDC) spoke about important trends in the area of data centres.
CDC’s two data centre campuses, connected by a data fibre, host the vast majority of the Australian federal government’s computing capacity, and increasingly the state governments’ computing capacity as well.
Two years ago, energy efficiency was the big focus. Now, a lot of the government’s infrastructure is now housed within highly efficient data centres that were purpose built for that outcome. So, what is the next thing?
Mr. Boorer said that the worldwide explosion of data has significantly altered the thinking around data centres.
“When organisations go to a more cloud-centric environment, whether public or private cloud, they have a 10-fold increase in network overhead. Combine that with the fact that the Australian government predicts, based on their historical data, that their data that they will have to manage and work with, will increase 100-fold in the next decade,” Mr. Boorer said highlighting the scale on ongoing change.
In this environment, the government is looking to adopt an ecosystem perspective. Public sector organisations are being brought together, as there is a huge amount of inter-agency information-sharing. Then third-party service providers are being brought inside the secure tent of this ecosystem.
Data centres are being consolidated and agencies are being encouraged to move in. This reduces the cost of moving data around, deals with performance and latency issues and improves security. It is easier and more cost-effective to secure aggregated data, compared to data in multiple locations.
That kind of density of government agencies also creates opportunities for data lakes and sharing of computing resources, where computing can be re-directed towards whichever agencies have high requirements at any point of time, thereby improving resource utilisation.
The digital transformation journey in many ways is underpinned by the infrastructure layer, from on-premise to co-location to hybrid or private clouds or public clouds. Mr. Boorer said, "Normally, you would get all of them from different locations, with their associated costs and overheads. If you can create an ecosystem and have all of those entities in the same physical location, then suddenly the overall cost of the digital transformation, the time it takes to achieve and the outcomes can be significantly enhanced."
The final trend is that workloads are moving towards the point of data aggregation. More and more big cloud providers are moving closer to where the data resides.
Polling questions and discussion
An overwhelming 86% of the delegates said that the greatest impact of their data centre being down would be on public service and their image. Around 64% replied that they have tools for tracking the efficiency of their data centre, while 18% are outsourcing the tracking to a third party.
Mr. Joshua Au, Head of Data Centre at the Agency For Science, Technology and Research (A*STAR) said that the benefit of having a product is that it provides visibility to many factors and metrics through the right instrumentation. And getting visibility is the first step towards improving efficiency.
Power Usage Effectiveness or PUE, a metric developed by a consortium called The Green Grid, is the most popular measure of a data centre’s energy efficiency. It is the ratio of total energy consumed within a facility to the energy consumed by the IT equipment. An ideal PUE is 1.0.
But the PUE rating has limitations. As Mr. Boorer pointed out, “The PUE rating is a big blunt instrument. You might have an excellent PUE rating but you might have lots of racks of equipment which people aren’t even using. They have been sitting there for 5 years, burning your electricity, which is good for your PUE, but it doesn’t deliver any sort of service.”
Mr. Mark Tan, IT Consultant, Server Infrastructure at the Ministry of Culture, Community and Youth (MCCY), brought up the opposite scenario of having new, empty data centres, large scale enough to be tier-3 or even tier-2 with racks and UPS, but no computing devices. Because there is no computing equipment inside, PUE is very high.
Mr. Au said that there is a need to go back to first principles and ask why is the PUE being measured. PUE is not meant to compare site A with site B. No two businesses are the same and no two sites are not the same. Even for the same site, if you calculate within 24 hours, the answer will change, because IT load changes. Temperature, humidity also change. The purpose of PUE is to act as a point of reference, to allow data centre managers to start reflecting on energy efficiency.
In view of the inadequacies of PUE, last year The Green Grid introduced the ‘Performance Indicator’ metric in a new white paper, which incorporates two more factors in addition to PUE: 1) IT Thermal Conformance (acceptable IT temperatures during normal operation); and 2) IT Thermal Resilience (acceptable IT temperatures during cooling failure or maintenance).
Resiliency vs efficiency
The conundrum here is somewhat similar to the security vs convenience dilemma faced in the area of cybersecurity.
Efficiency could be improved by compromising on resiliency, on things like Building Management Systems (BMS), fire suppression system, by turning off the UPS (uninterruptible power supply).
Who would want pure efficiency without resiliency or pure resiliency with low efficiency and astronomically high costs? Each business would have to find their own answers on what is the right balance for them, between the two.
Applying standards to small data centres – how do you define a data centre?
The Singapore Standard SS 564 Green Data Centres - Energy and environmental management systems was first published in January 2011, by the erstwhile Infocomm Development Authority or IDA (now Infocomm Media Development Authority or IMDA). It helps organisations establish systems and processes necessary to improve the energy efficiency of their data centres. There have been subsequent revisions and additions. In addition, IDA and Building and Construction Authority (BCA) jointly developed an additional Green Mark rating system for new and existing data centres. It takes into account energy efficiency, water efficiency, sustainable construction and management, indoor environment quality and other green features.
But several government agencies are now drawing on centralised computing resources and only have small data centres within the agency, with very few racks. Sometimes, these small data centres are distributed across locations driven by business needs. For some of these agencies, IT is a support function, though IT loads are increasing for all agencies, regardless of whether they are an IT-focused organisation or not.
But compliance with standards can be an issue for them. The investment required to gain visibility of the pertinent factors and work on them can be too high. It might not make sense from a cost-benefit perspective, not generating adequate ROI or not generate ROI soon enough.
Mr. Ngoh Lek Heng, Executive Manager (Green Data Centre), IMDA said that there are other considerations in addition to Green Mark, for choosing a data centre to house equipment.
The consensus appeared to be that Green Mark is a great starting point for the conversation and it provides benchmarks to work towards, though it might not be enough on its own.
Some participants asked if they would all need to get instrumentation from vendors for monitoring or if it was possible to benchmark once and then use a calculated method on a regular basis to review whether the standards were being met. If the standards are not being met or if they change, then they could get outside help. (It is important to note that 70% of the agencies represented at the session said that they are currently seeking a consultant/ expert for data centre energy efficiency assessment. Around 38% responded that they are currently monitoring energy efficiency regularly with in-house expertise, while for 50%, it is managed by an external partner.)
The other related question which came up was how to define a data centre. Currently any room which serves computing and networking and has data equipment is considered to be a data centre. But should the same standards be applicable to the small rooms mentioned above, as to large, industrial scale data centres with maybe hundreds or thousands of racks? And how should those small data centres comply with standards, in view of the possibly unfavourable cost-benefit analysis.
Efficiency and data centre location
When asked about the advantages of locating data centres in cold climates, where cooling costs might be lower, Mr. Boorer gave the example of data centres in Norway, built in stone mountains that used to be NATO ammunition storage. Water from 80 m below sea level is used for cooling. Cooling basically costs nothing. But the latency in the fibre connectivity from there to London or Frankfurt would render it useless for high performance computing or for banks and financial service companies, which would prefer to be right next to a stock exchange if possible.
There is a trade-off between the applicability of the resource relative to the business vs energy efficiency.
In addition, if for sovereignty and performance reasons, the data centre needs to be inside your building, then a data centre in Norway would be of no use. Mr. Boorer said that though Singapore has disadvantages of temperature and humidity, there is the potential of finding numerous small improvements, adding up to significant gains in efficiency. Obtaining each of those small gains requires experience, knowledge and knowhow.
Optimise according to business needs
The conclusion was the optimisation needs to happen according to the specific business needs of each organisation. That raises the question as to what is ‘optimisation’ in the context of data centres. Safety margins have to be maintained. Then there are the cost-benefit considerations. But at the same time, bold steps might be required and boundaries found which can be pushed. According to Mr. Ngoh, that is what IMDA is trying to do.